METHOD OF MAKING A CHEMICAL-MECHANICAL POLISHING SLURRY AND THE POLISHING SLURRY
FIELD OF THE INVENTION
A polishing slurry composition and its method of making for planarization of silicon semiconductor wafers by mechanical polishing of the wafer. More particularly,
a composition for polishing a wafer having tungsten lines and vias through silicon dioxide dielectric layers where the slurry has a high removal rate selectivity for the tungsten in relation to the silicon dioxide removal rate.
BACKGROUND OF THE INVENTION
Integrated circuit complexity has continued to evolve placing increasingly more demanding speci ications on the processes used in their manufacture. As the requirement for increasing the density of active devices on an individual chip has escalated, the requirement for greater flatness, over long distance and short distance, on the surfaces, top and bottom, of the wafer has also evolved. A flat surface is desirable for several reasons. Flatness is a requirement for cooperation with the optical focusing characteristics of optical stepper devices. As the optical lens requirements for increased resolution has increased, the depth of field of the lens has decreased. Also, attachment of the interconnection metallization to their underlayer is improved if the metal is not required to pass over abrupt underlying steps. In addition, flatness improves ability to fill via holes and lines through apertures in the dielectric.
Various processes have been used for planarization. One such process known as chemical-mechanical polishing (CMP) is presently being used in the most demanding
applications. CMP enhances the removal of surface material over large distances and short distances by simultaneously abrading the surface while a chemical etchant
selectively attacks the surface.
It is known from the pπor CMP efforts that for a particular CMP slurry to be successful it must exhibit significantly different etch removal rates for at least two different materials on the wafer surface. This is called selectivity and is normally shown as a different etch rate for the metal vis-a-vis the interlayer dielectric.
CMP is becoming a preferred method of planarizing tungsten interconnects, vias and contacts. With proper process parameters, CMP tungsten processing has shown significantly improved process windows and defect levels over standard tungsten dry etch-back processing. One significant advantage of CMP tungsten processing is that it has a highly selective polish rate for tungsten as compared to the dielectric. This selectivity allows for over-polishing while still achieving a flat tungsten
stud. When overetching using etch-back technology, the contact or via becomes further recessed which creates a serious disadvantage since overetching is frequently required to remove defects. The advantage of CMP, however, can be offset by the creation of significant levels of defects during polishing, such as scratches.
Accordingly, the success of tungsten CMP processing performance is linked to improvements in selectivity and defect control.
For various reasons, prior CMP slurries have not been as effective as needed. Most significantly, deep or wide scratch defects of the underlying surface by the abrasive have caused problems. Also, since a large volume of CMP slurry is required, the development of a low-cost chemical composition is essential. SUMMARY OF THE INVENTION
It is an object of this invention to provide a practical tungsten CMP slurry having excellent selectivity which enables planarization of a silicon integrated circuit
wafer with a low-defect count and excellent uniformity. These objects are achieved by formulating the slurry from abrasive particles having a sufficiently small median diameter and a very tight diameter variation and by thoroughly prcmixing the abrasive particles with a surfactant suspension agent before combining the oxidizer. A feature of this invention is that it provides a stable polish rate over a wide
range of solid slurry content so that the process is more easily controlled. BRIEF DESCRIPTION OF THE DR WINGS
FIG. 1 is a plot of W removal rate for various wt% alumina in slurry with a fixed oxidizer percentage Fe(NO3)3 • 9H,O = 5 wt%. FIG. 2 is a normalized plot of removal rate for various wt% of Fe(NO3)3 •
9H,O for a fixed abrasive percentage Al2O3 = 2.8 wt%.
FIG. 3 is a chart depicting the average total defect count for slurries made by different mixing sequences.
FIG. 4 is a chart depicting the oxide removal rate and tungsten selectivity for different slurry batches.
FIG. 5 is chart of tungsten removal rate and uniformity for different ferric salt oxidizers.
FIG. 6 is a graph of the general shape of the change in defect count as a function of aging for the slurry of this invention.
DETAILED DESCRIPTION OF THE INVENTION
Commercially available CMP equipment and sluπy materials are currently available for planarization of integrated circuits with tungsten (W) vias through silicon dioxide dielectric layers. These commercial slurry materials each exhibit problems such
as high scratch counts, high material cost, rapid settling and poor tungsten selectivity.
We have developed a sluπy and a method for making the slurry which substantially addresses and reduces these problems.
Our slurry comprises abrasive particles of a selected diameter, a ferric salt oxidizer and a suspension agent. We have also discovered that it is beneficial to follow an order of adding and of mixing the slurry components for optimum results. The suspension agent should be mixed thoroughly with the abrasive particle aqueous concentrate before adding diluted oxidizer up to the final volume.
We have also discovered a previously unobserved effect. Specifically, we find
that the planarization results can be still further improved if the completely mixed slurry is allowed to age for one day or more before being used in that scratch count is still further diminished. While one day of aging improves the scratch results, significant further improvement is obtained with longer aging of more than three days
and preferably more than five days.
The abrasive particles can be any of the commonly used abrasives such as alumina ( Al2O3), silicon carbide (SiC), Ceria (CeO2), silicon nitride (Si3N4) and silicon dioxide (SiO2). However, we have discovered that a small particle median diameter shows advantages and is preferred and that the distribution from the preferred diameter median should preferably be very tight in order to obtain reduced scratch count. The median diameter of the prior art slurry particles are 0.400 microns, but for our sluπy we prefer a median particle size of less than 0.4 microns, preferably 0.220 microns.
The preferred distribution of particle size is much tighter than in the prior art also. Our preferred distribution is a one sigma deviation of 0.050 microns or less. We have determined through experimentation that both of these dimensions are important to
achieve low incidence of scratching. Using standard optical measurement equipment,
generally, the particle sizes used in prior CMP slurries were in the range of 0.4 to 0.7 micron diameter or larger with little attention to the tightness of the distribution.
The preferred oxidizer is a ferric salt, selected from the group consisting of Fe(NO3)3 • 9H,O, FeCl3 • 6H2O, Fej(SO4)3 • 5H2O, and FeNH4(SO4)2 • 12H2O. The suspension agent is preferably an aqueous based surfactant to improve the colloidal behavior of the abrasive particles in the H2O system. For the purposes of this
invention a commercially available aqueous mixture of propyleneglycol and methylparaben was used as the suspension agent. The suspension agent can also be formulated from the following classes: 1 ) glycols such as ethylene glycol, propylene glycol and glycerol;
2) polyethcrs such as polyethylene glycol;
3) aliphatic poly ethers;
4) alkyl sulfates;
5) alkoxylated alkyphenols. We believe that CMP of tungsten films takes place by ( 1 ) a chemical oxidation of the tungsten surface with a suitable oxidizing agent in an aqueous solution, (2) followed by mechanical abrasion of the more brittle metal oxide which had formed on the surface by the solid abrasive particles present in the aqueous suspension. Both the oxidation and abrasion continue simultaneously and continuously. The reaction for oxidation of tungsten by the ferric ion is
W+ 6Fe + 3 H2O - WO3 + 6Fe + 6H*
and occurs in an acid solution.
The oxidizer component has other members of the class. Representative
members of the class include suitable Fe (III) compounds, such as the following:
Ferric chloride hexahydrate, FeCl, • 6H,O
Ferric sulfatc pentahydrate. Fe2(SO )3 • 5H,O and Ferric ammonium sulfate dodecahydrate. FeNH4(SO4); ■ 12H:O Experiment 1 A pair of CMP experiments (Experiment 1 and Experiment 2) as set forth below were performed to determine the relationship between the percentage of alumina
solid content and percentage of ferric nitrate content in a slurry and the tungsten removal rate. No surfactant was employed in Experiment 1. In this experiment, the oxidizer was ferric nitrate Fe(NO3)3 • 9H:O, which was held constant at 5 weight percent (5 wt ). The results of various percentages of abrasion particles of Alumina are plotted against the tungsten removal rate in FIG. 1. The experimental procedure
and equipment for Experiment 1 were as follows:
Equipment: Strasbaugh model 6DS-SP polisher.
Wafer carrier Standard design. Polish Pad: Industry standard. Test Monitor Pre-measured blanket tungsten films on Ti/TiN barriers over plasma TEOS oxide to monitor W removal rate and within wafer non-uniformity.
Dummy Wafers: Tungsten Dummy wafers identical to test monitors described above, except for the W thickness pre- measurement. Oxide dummv wafers coated with 2500 A
oxide.
Metrology tool: Prometrix RS-55. Pre and post polish measurements taken on 49 sites across each wafer with 9mm edge exclusion. Slurry agitation: Constant agitation during testing. PROCESS PARAMETERS
Bum in: New polish pads are burned in with 2 cycles of 120
seconds polish time/cycle using oxide dummy wafers. Process Cycle: variable time, 5 psi, 25 rpm carrier, 100 rpm table, 150 ml/min sluπy flow. Loading Sequence: For each of seven slurry mixture text compositions selected, perform polish test on 8 tungsten dummy wafers for 220 sec and 4 tungsten Test Monitor wafers for 60 seconds. Test Compositions: Al2O3 -- 1 wt%, 2 wt%, 3 wt%, 4 wt%, 5 wt%, 6 wt% and 7 wt%. The Fe(NO3)3 • 9H2O is fixed at 5 wt% for all slurries used in Experiment 1. This experiment, as shown in FIG. 1, has a nearly constant W removal rate of 450θA/min — 490θA/min from 2-6 wt% of abrasion particles of Al,O3. Below 2 wt%
the removal rate is substantially reduced. In the flat region, the removal rate is essentially independent of the alumina concentration. This suggests that the removal rate depends more on the oxidation rate of the elemental tungsten near the surface and that at these wt% of particles the tungsten oxide is being removed about as fast as its being formed. It is not understood why the curve shows a removal rate drop-off at the percentages above 6 wt%.
The prior CMP tungsten polishing publications teach that the polishing rate is linearly proportional to the concentration of the alumina particles in the concentration range of 3 to 7 wt. percent. However, our experiment shows that the lower
concentration can sustain the high polishing rate if the alumina particle is appropriately sized and has the proper size distribution. It appears that certain particle sizes contribute more to polishing. Larger particles when mixed with smaller particles may hinder the polishing performance of the smaller particles by maintaining too large a distance between the wafer and the polishing pad. Experiment 2
For Experiment 2, the same experimental procedure and equipment were used as set forth in Experiment 1 except that the Test Points for this experiment held the alumina at 2.8 wt% and used six points of different weight percent of ferric nitrate, i.e., 3 wt%, 4 wt%, 5 wt%, 6 wt%, 7 wt% and 9 wt%.
FIG. 2 for Experiment 2 shows the results of W removal rate plotted normalized against removal rate for 5 wt% Fe(NO3)3 • 9H2O. The results show a strong W removal rate dependence on concentration of Fe3+ between 3 wt% to 5 wt% ferric nitrate. At increased Fe3+ concentration, i.e., 5 wt% to 9 wt%, the removal rate is nearly constant for a fixed 2.8 wt% alumina concentration.
This behavior suggests that below 5 wt% ferric nitrate concentration that the polish rate is controlled by the chemical oxidation rate of the freshly exposed W surface undergoing continuous abrasion by Al2O3 particles. Above 5 wt% ferric nitrate, the increased Fe + concentration does not lead to a higher polish rate, which
suggests that surface WO- abrasion rate becomes the rate limiting mechanism.
Experiment 3
Addition of a surfactant not only improved the colloidal suspension time but the surfactant also had an unexpected, very significant effect on the defect scratch count level. The instrumentation counted all scratches of 0.3 microns and above. The
surfactant also unexpectedly affected the W selectivity. For Experiment 3 the
surfactant used was a commercially available aqueous mixture of propylene glycol and methyl paraben from Universal Photonics Inc., sold under trade name EVERFLO. The order of mixing the slurry components also had an unexpected effect on the results.
A. In Experiment 3, three different 2.5 gallon slurry batches were prepared, each batch having exactly identical ingredient ratios but were combined in a different sequence or manner of adding in the suspension agent:
Batch A: Surfactant was added to container holding alumina and ferric nitrate, previously diluted to near the final volume. Batch B: Surfactant was added to alumina concentrate. After brief stirring to homogenize the mixture, diluted ferric nitrate was added up to the final volume. Batch C: Surfactant added to alumina concentrate and mixed by a magnetically driven stirrer for 2 hours, then the diluted oxidizer, ferric nitride, was added to make up the final batch volume. B. Sluπy Composition
AI2O3 concentrate @ 14 wt% Al2O3 solids: 20% of total sluπy volume or Al2O3 solids at 2.8 wt% of total slurry weight. Surfactant: 15% of total slurry volume Ferric nitrate: 5 wt% of total slurry weight
D.I. water 65% of total slurry volume
C. Equipment
Defect measurement: ESTEK, model 8500 Oxide removal and non-uniformity: Prometrix SM200 Tungsten removal and wafer non-uniformity: Prometrix RS55
Polisher Strasbaugh 6DS-SP Pads: Industry standard
D. Procedure
Polishing in Strasbaugh carried out at 5 psi spindle down force, 25 rpm spindle rotation, 100 rpm table rotation.
A polishing pad was wet-idled overnight. The pad was pre-
soaked with sluπy for 2 min. Dummies with = 3000 angstroms of tungsten film were polished for 120 seconds, a total of two cycles for pad burn in. Next, two rate monitors with = 8000A of tungsten were polished for 60 seconds in one cycle. Then, two oxide-layered wafers were polished for 60 seconds to determine oxide removal rate. Lastly, two additional oxide wafers were polished for 10 seconds at 5 psi and 30 seconds at 2 psi for defect evaluation. Both oxide rate and defect monitor wafers had approximately 2500 A PETOS. The polish pad was rinsed with DI water thoroughly between different slurry batches.
FIG. 3 shows that when performing CMP using no surfactant that
approximately a total of 2000 defects of 0.3 microns or above are counted. For Batch
A, where surfactant was added after diluting the abrasive and Fe(NO3)3 to near final volume, the data shows that the defects dropped only slightly to approximately 1800.
A dramatic decrease in scratching to 400 counts is obtained with Batch B where
surfactant is added before the dilution and oxidizer with only minimal mixing. The result with Batch C employing a thorough pre-mixing of the abrasive particles with the surfactant before dilution and oxidizer added was seen to reduce the defect count to less than 200. It is believed that the thorough mixing of the surfactant prior to addition
of the oxidizer allows the suspension agent to engage and completely coat each alumina panicle by nature of the organic surfactant molecules making the particle surface non-polar. It is believed that this precludes the particle from acquiring a charge from the oxidizer. thereby preventing agglomeration. Accordingly, when the
mechanical action of the particles are abraded across the W surface, the slurry particles remain smaller and cause less gouging. As seen in FIG. 4, the oxide removal rate
(A/min) is also significantly different for each batch resulting in a very different selectivity which increases from 140: 1 to over 400: 1 depending on the mixing sequence. The prior art teaches that tungsten to TEOS polish rate selectivity for CMP is less than 180, see D. Scherber and M. Neville, Chemical and Mechanical Aspects of Consummables Used in CMP- Semicon Southwest Proceedings, p. 36-60 (1994). Further, the known .art teaches that high tungsten/TEOS selectivity requires reduced tungsten polish removal rate. As indicated above, our invention teaches that both of these limitations can be avoided.
Experiment 4
For Experiment 4, slurries were made using the same process as described for making Batch C in Experiment 3 for several ferric (Fe3+) salts and the tungsten polish
rates and uniformity were measured.
The standard Strasbaugh W-polish configuration of Experiment 1 was employed. Each new slurry, after thorough rinsing, was used to saturate the polishing pad and two bura-in wafers were polished followed by two tungsten test wafers. The results were as plotted in FIG. 5. It is clear that each *f »he four ferric salts can be successfully used to formulate the W polishing slurry. The ferric chloride displays a polishing rate similar to ferric nitrate, while the sulfate salt polishes at a higher rate, but
with somewhat poor uniformity. AGING TESTS
Using the same procedure for making Batch C of the slurry using Fe(NO3)3 •
9H2O as described above for Experiment 3, and using the same test procedure for measuring defect counts (scratching), the effects of aging the slurry were measured. Two Test Wafers were run and data obtained in one-day increments over a six-day
aging period. As seen from FIG. 6, the scratch count decreases dramatically over the six-day aging period. After six days aging, the decrease defect count is decreased to less than 10% from the one-day defect count value.
It will be apparent to those skilled in the art having reference to this disclosure that other modifications of this invention beyond those embodiments specifically described here may be made without departing from the spirit of the invention. Accordingly, it is not our intention to restrict our invention to the embodiments described but that the scope of the inventions shall be determined by the following
claims. With this in view,